GPT Realtime 2.1
OpenAI · Live conversation · openai/gpt-realtime-2.1
Live conversation streaming
The most capable speech-to-speech: tools, interruption, tone. Also the most expensive per minute.
What it costs
audio tokens: $32/1M in, $64/1M out (~10 tokens/s of audio).
The per-minute figure converts audio tokens at the supplier's own rate (10/10 a second in and out). Prices are ballpark, pay-as-you-go, mid-tier — they move constantly. Last checked 2026-07-29.
How it performs
A blank cell means the row does not state it, not that the answer is no. Word-error rates come from one independent harness, so they compare with each other only.
Using it
Runs on the supplier's API, on your own key. You bring the supplier's key and pay their rates directly.
These are not AI-token prices, and these engines do not answer at the hosted endpoints — 00 speaks to them from your machine, with your credentials. The speech models that do run on AI tokens are one click away on the media page.
Pick it per agent in the app. Get 00.
Subscribe to new models